Belief-based Hybrid Argumentation for Threat Assessment

This paper describes a mixed-initiative model of knowledge discovery capable of monitoring a dynamic environment, in which uncertain and unreliable messages can be reasoned over for recognizing human activities and predicting likely threats. The model represents "an argument assistant" helping an analyst in argument production by considering pro and contra arguments from uncertain transient information while seeing each piece of this information as an element of alternative stories (hypotheses based on "what might happen"). These hypotheses are evaluated within the framework of the Transferable Belief Model by assigning beliefs to each argument, combining these beliefs, and selecting a story (hypothesis) based on the highest pignistic probability. Anytime decision making provides decision quality control by weighing time and hypothesis credibility.